Reproduce “Measuring Empowerment from Offline Data for Intrinsically Motivated Exploration” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ishir Garg, Vivek Myers, Qiyang Li, Catherine Ji, Benjamin Eysenbach, Sergey Levine
The method estimates maximal empowerment from unlabeled offline trajectories and uses the map to guide online exploration in sparse-reward, goal-conditioned tasks. It learns future-state occupancies for different skills from offline transitions, favoring states with varied futures. The authors report faster learning and higher success than baseline CRL on five task–dataset pairs, with on-par performance and no reported harm on other tested environments; downstream online interaction remains necessary.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.